Comparison of Cox Regression and Parametric Models for Survival of Breast Cancer Patients with 1-3 Positive Lymph Nodes

Abstract

The purpose of this study was to compare the performance of Cox regression and two parametric models under weibull and log-logistic distributions by applying with 90 breast cancer patients with 1-3 positive lymph nodes treated at the Faculty of Medicine, Chiang Mai University from 2001 to 2007. The independent variables to be studied were as follows: tumor grade, tumor size, number of examined nodes, menopause, estrogen receptor, progesterone receptor, radiotherapy, chemotherapy, regimen and endocrine therapy. The Akaike information criterion (AIC) was used for comparing model efficiency. The study results of univariate analysis with Cox regression and parametric models indicated tumor size, radiotherapy, and endocrine therapy were statistically significant effect on survival time. For multivariate analysis, it showed that tumor size was the statistically significant factor on survival time for both parametric models, under weibull and log-logistic distributions, while the endocrine therapy was the statistically significant effect on survival time for Cox regression and the parametric model under weibull distribution. Based on AIC, the Cox regression model was the best appropriate model with the smallest value of AIC. Keywords :  Cox regression model, Parametric model, Weibull distribution, Log-logistic distribution